Example top eQTLs: T statistics in 44 tissues

Method Outline

Multi-tissue eQTL mapping is basically trying to cluster these (noisy!) profiles

Tissue-strong eQTLs: Testis

Tissue-strong eQTLs: Whole-Blood

Tissue-strong eQTLs: LCLs

Also many “tissue-consistent” eQTLs

Method Outline

• Model true \(\beta\) values as mixture of “tissue-specific” and “tissue-consistent” (“BMAlite”)
• …plus additional components to capture other sub-types of QTL
• Fit the model; compute posterior distribution on effects in each tissue.
• Declare eQTL as ``significant" in tissue if data+model suggest strong confidence in direction (sign) of effect.

That U is still with us…

But tissue-specific dominated by a few tissues…

Lots of sharing, but similar tissues share more

Maybe some false positives?

Tissue-specific eQTLs don’t (usually) show Tissue-specific expression

• log(mean Expression) for genes with Testis-specific eQTLs

Things to worry about: Correlations between tissues

Summary

• We have (very) preliminary results multi-tissue mapping for 44 tissues
• Overall, continue to see some strong shared eQTLs as well as some highly tissue-specific signals in a few tissues
• And lots in between (but beware of thresholding effects…)
• Strong tissue-specific eQTLs not primarily driven by tissue-specific expression
• Brain tissues may show less overlap with others (and more with one another)

(Thanks to Karl Broman and his R/qtlcharts package)